def ddf_mul_resnet50(pretrained=False, **kwargs):
model_args = dict(block=DDFMulBottleneck, layers=[3, 4, 6, 3], **kwargs)
return build_model_with_cfg(ResNet,
'ddf_mul_resnet50',
default_cfg=default_cfgs['ddf_mul_resnet50'],
pretrained=pretrained,
**model_args)
def _create_regnet(variant, pretrained, **kwargs):
return build_model_with_cfg(model_cls=RegNet,
variant=variant,
pretrained=pretrained,
default_cfg=default_cfgs[variant],
model_cfg=model_cfgs[variant],
**kwargs)
def ddf_add_resnet152(pretrained=False, **kwargs):
model_args = dict(block=DDFAddBottleneck, layers=[3, 8, 36, 3], **kwargs)
return build_model_with_cfg(ResNet,
'ddf_add_resnet152',
default_cfg=default_cfgs['ddf_add_resnet152'],
pretrained=pretrained,
**model_args)
def _create_vision_transformer(variant,
pretrained=False,
distilled=False,
**kwargs):
default_cfg = deepcopy(default_cfgs[variant])
overlay_external_default_cfg(default_cfg, kwargs)
default_num_classes = default_cfg['num_classes']
default_img_size = default_cfg['input_size'][1:]
num_classes = kwargs.pop('num_classes', default_num_classes)
img_size = kwargs.pop('img_size', default_img_size)
repr_size = kwargs.pop('representation_size', None)
if repr_size is not None and num_classes != default_num_classes:
# Remove representation layer if fine-tuning. This may not always be the desired action,
# but I feel better than doing nothing by default for fine-tuning. Perhaps a better interface?
repr_size = None
if kwargs.get('features_only', None):
raise RuntimeError(
'features_only not implemented for Vision Transformer models.')
model_cls = VisionTransformer
model = build_model_with_cfg(model_cls,
variant,
pretrained,
default_cfg=default_cfg,
img_size=img_size,
num_classes=num_classes,
representation_size=repr_size,
pretrained_filter_fn=checkpoint_filter_fn,
**kwargs)
return model
def _create_rexnet(variant, pretrained, **kwargs):
feature_cfg = dict(flatten_sequential=True)
if kwargs.get('feature_location', '') == 'expansion':
feature_cfg['feature_cls'] = 'hook'
return build_model_with_cfg(ReXNetV1,
variant,
pretrained,
default_cfg=default_cfgs[variant],
feature_cfg=feature_cfg,
**kwargs)
def _create_tnt(variant, pretrained=False, **kwargs):
if kwargs.get('features_only', None):
raise RuntimeError('features_only not implemented for Vision Transformer models.')
model = build_model_with_cfg(
TNT, variant, pretrained,
default_cfg=default_cfgs[variant],
pretrained_filter_fn=checkpoint_filter_fn,
**kwargs)
return model
def _create_normfreenet(variant, pretrained=False, **kwargs):
print('create nf net with pretraiend=' + str(pretrained))
model_cfg = model_cfgs[variant]
feature_cfg = dict(flatten_sequential=True)
return build_model_with_cfg(
NormFreeNet, variant, pretrained,
default_cfg=default_cfgs[variant],
model_cfg=model_cfg,
feature_cfg=feature_cfg,
**kwargs)
def scearesnet18(pretrained=False, **kwargs):
model_args = dict(block=BasicBlock,
layers=[2, 2, 2, 2],
block_args=dict(attn_layer=SCEAModule),
**kwargs)
default_cfg = timm_resnet_cfg()
return build_model_with_cfg(ResNet,
'scearesnet18',
default_cfg=default_cfg,
pretrained=pretrained,
**model_args)
def scseresnet18(pretrained=False, **kwargs):
model_args = dict(block=BasicBlock,
layers=[2, 2, 2, 2],
block_args=dict(attn_layer=SCSEModule),
**kwargs)
default_cfg = timm_resnet_cfg(url='', interpolation='bicubic'),
return build_model_with_cfg(ResNet,
'scseresnet18',
default_cfg=default_cfg,
pretrained=pretrained,
**model_args)
def _create_mnv3(model_kwargs, variant, pretrained=False):
if model_kwargs.pop('features_only', False):
load_strict = False
model_kwargs.pop('num_classes', 0)
model_kwargs.pop('num_features', 0)
model_kwargs.pop('head_conv', None)
model_kwargs.pop('head_bias', None)
model_cls = MobileNetV3Features
else:
load_strict = True
model_cls = MobileNetV3
return build_model_with_cfg(model_cls,
variant,
pretrained,
default_cfg=default_cfgs[variant],
pretrained_strict=load_strict,
**model_kwargs)
def _create_vision_transformer(variant, pretrained=False, **kwargs):
default_cfg = deepcopy(default_cfgs[variant])
overlay_external_default_cfg(default_cfg, kwargs)
default_num_classes = default_cfg['num_classes']
num_classes = kwargs.pop('num_classes', default_num_classes)
if kwargs.get('features_only', None):
raise RuntimeError('features_only not implemented for Hierarchical Visual Transformer models.')
model = build_model_with_cfg(
VisionTransformer,
variant,
pretrained,
default_cfg=default_cfg,
num_classes=num_classes,
**kwargs)
return model
def _create_volo(variant, pretrained=False, **kwargs):
if kwargs.get('features_only', None):
raise RuntimeError(
'features_only not implemented for Vision Transformer models.')
return build_model_with_cfg(VOLO, variant, pretrained, **kwargs)
